Potential Predictability of Regional Temperature Extremes Estimated from High-Resolution Large Ensemble Simulations

Extreme weather events such as heavy rainfall and heat waves are crucial issues. However, seasonal-to-interannual prediction of such local extremes remains challenging, particularly in the mid-latitudes, due to a small signal-to-noise ratio and the limited resolution of climate models in representing mesoscale circulation and complex topography.

In this study, we assess the potential predictability of both the mean state (monthly mean temperature) and extremes (monthly maximum temperature and the number of extremely hot days) over Japan. We employ 100-member large-ensemble AGCM simulations and high-resolution regional climate model downscaling. Two types of simulations are analyzed: historical simulations, forced by historical SST, sea ice, and atmospheric forcings (ozone, greenhouse gases, and aerosols); and non-warming simulations, forced by detrended SST and preindustrial levels of these forcings.

Our results indicate that the potential predictability of temperature extremes is generally lower than that of the mean state. Notably, we detected a significant and spatially localized difference in predictability of extremes between historical and non-warming conditions, whereas no such difference was found for the mean state.
To elucidate the mechanisms underlying this difference in predictability, we further examine variability in several ocean basins and their teleconnections to local atmospheric circulation.